代码合并

.gitignore

@@ -1,7 +1,7 @@
 *.un~
 *.pyc
 *.swp
-# *.json
+*.json
 res/
 .DS_Store
 
@@ -189,6 +189,7 @@ pip-selfcheck.json
 
 .vscode/
 
+<<<<<<< HEAD
 
 Data/ml-20m
 Data/ml-10m
@@ -202,3 +203,7 @@ experiments/
 !*.json
 *.nohup
 Data/preprocessed
+=======
+Data/
+experiments/
+>>>>>>> f66f2534ebfd937778c7174b5f9f216efdebe5de

dataloaders/ae.py

@@ -91,7 +91,6 @@ def __init__(self, user2items, item_count):
         for user, useritem in enumerate(user2items.values()):
             for _ in range(len(useritem)):
                 user_row.append(user)
-
         # Column indices for sparse matrix
         item_col = []
         for useritem in user2items.values():
@@ -102,6 +101,7 @@ def __init__(self, user2items, item_count):
         sparse_data = sparse.csr_matrix((np.ones(len(user_row)), (user_row, item_col)),
                                         dtype='float64', shape=(len(user2items), item_count))
 
+
         # Convert to torch tensor
         self.data = torch.FloatTensor(sparse_data.toarray())
 
@@ -127,7 +127,6 @@ def __init__(self, user2items, item_count):
             for _ in range(len(label_items)):
                 label_user_row.append(user)
         input_user_row, label_user_row = np.array(input_user_row), np.array(label_user_row)
-
         # Column indices for sparse matrix
         input_item_col = np.hstack(input_list)
         label_item_col = np.hstack(label_list)
@@ -142,6 +141,7 @@ def __init__(self, user2items, item_count):
         self.input_data = torch.FloatTensor(sparse_input.toarray())
         self.label_data = torch.FloatTensor(sparse_label.toarray())
 
+
     def split_input_label_proportion(self, data, label_prop=0.2):
         input_list, label_list = [], []
 

dataloaders/base.py

@@ -14,7 +14,10 @@ def __init__(self, args, dataset):
         self.test = dataset['test']
         self.umap = dataset['umap']
         self.smap = dataset['smap']
+<<<<<<< HEAD
         self.top50 = dataset['top50']
+=======
+>>>>>>> f66f2534ebfd937778c7174b5f9f216efdebe5de
         self.user_count = len(self.umap)
         self.item_count = len(self.smap)
 

datasets/base.py

@@ -4,7 +4,6 @@
 import numpy as np
 import pandas as pd
 from tqdm import tqdm
-
 tqdm.pandas()
 
 from abc import *

templates.py

@@ -9,6 +9,7 @@ def set_template(args):
 
         # args.dataset_code = 'ml-' + input('Input 1 for ml-1m, 20 for ml-20m: ') + 'm'
         args.dataset_code = 'ml-1m'
+
         args.min_rating = 0 if args.dataset_code == 'ml-1m' else 4
         args.min_uc = 5
         args.min_sc = 0
@@ -25,6 +26,7 @@ def set_template(args):
         args.train_negative_sampling_seed = 0
         args.test_negative_sampler_code = 'random'
         args.test_negative_sample_size = 100 if args.dataset_code == 'ml-1m' else 500
+        args.test_negative_sample_size = 100
         args.test_negative_sampling_seed = 98765
 
         args.trainer_code = 'bert'
@@ -41,6 +43,7 @@ def set_template(args):
         args.best_metric = 'NDCG@10'
 
         args.model_code = 'bert_cnn'
+
         args.model_init_seed = 0
 
         args.bert_dropout = 0.1

trainers/__init__.py

@@ -1,13 +1,19 @@
 from .bert import BERTTrainer
 from .dae import DAETrainer
 from .vae import VAETrainer
+<<<<<<< HEAD
 from .ncf import NCFTrainer
+=======
+>>>>>>> f66f2534ebfd937778c7174b5f9f216efdebe5de
 
 
 TRAINERS = {
     BERTTrainer.code(): BERTTrainer,
     DAETrainer.code(): DAETrainer,
+<<<<<<< HEAD
     NCFTrainer.code(): NCFTrainer,
+=======
+>>>>>>> f66f2534ebfd937778c7174b5f9f216efdebe5de
     VAETrainer.code(): VAETrainer
 }
 

trainers/base.py

@@ -11,8 +11,11 @@
 import json
 from abc import *
 from pathlib import Path
+<<<<<<< HEAD
 import math
 import numpy as np
+=======
+>>>>>>> f66f2534ebfd937778c7174b5f9f216efdebe5de
 
 
 class AbstractTrainer(metaclass=ABCMeta):
@@ -97,15 +100,23 @@ def train_one_epoch(self, epoch, accum_iter):
 
             average_meter_set.update('loss', loss.item())
             tqdm_dataloader.set_description(
+<<<<<<< HEAD
                 'Epoch {}, loss {:.3f} '.format(epoch + 1, average_meter_set['loss'].avg))
+=======
+                'Epoch {}, loss {:.3f} '.format(epoch+1, average_meter_set['loss'].avg))
+>>>>>>> f66f2534ebfd937778c7174b5f9f216efdebe5de
 
             accum_iter += batch_size
 
             if self._needs_to_log(accum_iter):
                 tqdm_dataloader.set_description('Logging to Tensorboard')
                 log_data = {
                     'state_dict': (self._create_state_dict()),
+<<<<<<< HEAD
                     'epoch': epoch + 1,
+=======
+                    'epoch': epoch+1,
+>>>>>>> f66f2534ebfd937778c7174b5f9f216efdebe5de
                     'accum_iter': accum_iter,
                 }
                 log_data.update(average_meter_set.averages())
@@ -120,15 +131,23 @@ def validate(self, epoch, accum_iter):
         average_meter_set = AverageMeterSet()
 
         with torch.no_grad():
+<<<<<<< HEAD
             tqdm_dataloader = tqdm(self.test_loader)
+=======
+            tqdm_dataloader = tqdm(self.val_loader)
+>>>>>>> f66f2534ebfd937778c7174b5f9f216efdebe5de
             for batch_idx, batch in enumerate(tqdm_dataloader):
                 batch = [x.to(self.device) for x in batch]
 
                 metrics = self.calculate_metrics(batch)
 
                 for k, v in metrics.items():
                     average_meter_set.update(k, v)
+<<<<<<< HEAD
                 description_metrics = ['NDCG@%d' % k for k in self.metric_ks[:3]] + \
+=======
+                description_metrics = ['NDCG@%d' % k for k in self.metric_ks[:3]] +\
+>>>>>>> f66f2534ebfd937778c7174b5f9f216efdebe5de
                                       ['Recall@%d' % k for k in self.metric_ks[:3]]
                 description = 'Val: ' + ', '.join(s + ' {:.3f}' for s in description_metrics)
                 description = description.replace('NDCG', 'N').replace('Recall', 'R')
@@ -137,7 +156,11 @@ def validate(self, epoch, accum_iter):
 
             log_data = {
                 'state_dict': (self._create_state_dict()),
+<<<<<<< HEAD
                 'epoch': epoch + 1,
+=======
+                'epoch': epoch+1,
+>>>>>>> f66f2534ebfd937778c7174b5f9f216efdebe5de
                 'accum_iter': accum_iter,
             }
             log_data.update(average_meter_set.averages())
@@ -162,6 +185,7 @@ def test(self):
 
                 for k, v in metrics.items():
                     average_meter_set.update(k, v)
+<<<<<<< HEAD
                 description_metrics = ['NDCG@%d' % k for k in self.metric_ks[:3]] + \
                                       ['Recall@%d' % k for k in self.metric_ks[:3]]
                 description = 'Val: ' + ', '.join(s + ' {:.3f}' for s in description_metrics)
@@ -189,6 +213,9 @@ def test_top50(self, top50):
                 for k, v in metrics.items():
                     average_meter_set.update(k, v)
                 description_metrics = ['NDCG@%d' % k for k in self.metric_ks[:3]] + \
+=======
+                description_metrics = ['NDCG@%d' % k for k in self.metric_ks[:3]] +\
+>>>>>>> f66f2534ebfd937778c7174b5f9f216efdebe5de
                                       ['Recall@%d' % k for k in self.metric_ks[:3]]
                 description = 'Val: ' + ', '.join(s + ' {:.3f}' for s in description_metrics)
                 description = description.replace('NDCG', 'N').replace('Recall', 'R')
@@ -205,8 +232,12 @@ def _create_optimizer(self):
         if args.optimizer.lower() == 'adam':
             return optim.Adam(self.model.parameters(), lr=args.lr, weight_decay=args.weight_decay)
         elif args.optimizer.lower() == 'sgd':
+<<<<<<< HEAD
             return optim.SGD(self.model.parameters(), lr=args.lr, weight_decay=args.weight_decay,
                              momentum=args.momentum)
+=======
+            return optim.SGD(self.model.parameters(), lr=args.lr, weight_decay=args.weight_decay, momentum=args.momentum)
+>>>>>>> f66f2534ebfd937778c7174b5f9f216efdebe5de
         else:
             raise ValueError
 
@@ -238,6 +269,7 @@ def _create_state_dict(self):
 
     def _needs_to_log(self, accum_iter):
         return accum_iter % self.log_period_as_iter < self.args.train_batch_size and accum_iter != 0
+<<<<<<< HEAD
 
     def _evaluate(self, candidate, top50, ks):
 
@@ -257,3 +289,5 @@ def _evaluate(self, candidate, top50, ks):
             metrics['Recall@%d' % k], metrics['NDCG@%d' % k] = np.array(hit).mean(), np.array(ndcg).mean()
 
         return metrics
+=======
+>>>>>>> f66f2534ebfd937778c7174b5f9f216efdebe5de

trainers/vae.py

@@ -62,7 +62,11 @@ def calculate_metrics(self, batch):
 
         # Annealing beta
         if self.finding_best_beta:
+<<<<<<< HEAD
             if self.current_best_metric <= metrics[self.best_metric]:
+=======
+            if self.current_best_metric < metrics[self.best_metric]:
+>>>>>>> f66f2534ebfd937778c7174b5f9f216efdebe5de
                 self.current_best_metric = metrics[self.best_metric]
                 self.best_beta = self.__beta
 

utils.py

@@ -34,6 +34,7 @@ def create_experiment_export_folder(args):
 
 
 def get_name_of_experiment_path(experiment_dir, experiment_description, args):
+
     experiment_path = os.path.join(experiment_dir, (experiment_description + "_" + str(date.today())))
     idx = _get_experiment_index(experiment_path)
     experiment_path = experiment_path + "_" + str(idx)